28-05-2014, 01:03 PM
Low Resolution Single Neural Network Based Face Recognition
Low Resolution Single Neural.pdf (Size: 572.68 KB / Downloads: 9)
Abstract
This research paper deals with the implementation of
face recognition using neural network (recognition classifier) on low-
resolution images. The proposed system contains two parts,
preprocessing and face classification. The preprocessing part converts
original images into blurry image using average filter and equalizes
the histogram of those image (lighting normalization). The bi-cubic
interpolation function is applied onto equalized image to get resized
image. The resized image is actually low-resolution image providing
faster processing for training and testing. The preprocessed image
becomes the input to neural network classifier, which uses back-
propagation algorithm to recognize the familiar faces. The crux of
proposed algorithm is its beauty to use single neural network as
classifier, which produces straightforward approach towards face
recognition. The single neural network consists of three layers with
Log sigmoid, Hyperbolic tangent sigmoid and Linear transfer
function respectively. The training function, which is incorporated in
our work, is Gradient descent with momentum (adaptive learning
rate) back propagation
INTRODUCTION
HE way of matching faces with the stored set of images
(database) is known as face recognition. The lengthy
tenure of success and achievements has blessed human
nervous system with abilities that are absent in basic
computing or even in modern parallel computing e.g. highly
interconnection, adaptive nature, learning skills and
generalization etc. In present or in coming times systems
based upon biological neurons contain some of such
characteristics [1].
Human brain has numerous highly interconnected
biological neurons, which on some specific tasks can perform
faster than super computers. Humans start recognizing faces as
they start growing up but for computers it’s a cumbersome
task. So idea is to a imitate computer system which recognize
faces as human brain does. Neural Networks has been widely
used in pattern recognition applications and it has performed
effectively in face recognition paradigm [2].
PROPOSED TECHNIQUE
This work deals with recognition faces using low resolution
images through Neural Networks. Carrying out the same task
with higher dimension can cause overfitting and
computational complexity. Data in high dimensions contain
redundancies and irrelevant parameters, neural networks
require large networks to cope up training of such data [4]. In
the proposed technique, image resizing is applied to get an
edge in terms of time and memory utilization over the high
resolution images.
In order to accomplish face recognition task just single
neural network is incorporated. Implementation is divided into
two phases. The pre-processing phase and the Neural
processing phase. In Pre-processing phase time effective pre-
processing is performed in order to make image data best fit
for neural network input. This phase output images of low
resolution, which have the information required for
recognition.
RESULTS AND DISCUSSION
The algorithm is implemented using 2.4 GHZ Pentium 4
machine with Windows XP and MATLAB 7.0 as the
development tool.
The ORL database has 40 subjects with 10 images per
person. Two set of images are required, one for the training of
the neural network and another set of images upon which
testing is done. In this research technique each subject’s out of
10 images picked 5 images as training examples and 5 images
as testing set for the neural network. After the training session,
the trained network is tested upon the unseen images and also
upon which training has been done.
In first case, 20 subjects (200 images) are taken as input,
and after multiple iterations of neural network training, in
testing deduced 94.5% results. Similarly for 30 subjects (300
images) and for 40 subjects (400 images) its giving 93% and
90.25% results respectively as shown in table I. Whereas LR-
SNN-T, Low Resolution Single Neural Network Technique
denotes the implemented technique.
CONCLUSION AND FUTURE WORK
Face Recognition is basically a classification problem. In
the research work neural network using backpropagation has
been trained as a face classifier to recognize faces with time
effective pre-processing, which greatly increases the
performance of the network. By lowering the resolution and
using Single Neural Network for whole recognition task,
computational complexity has been reduced many times.
As a future prospect.